Computer vision: how AI "sees" and understands images

Computer vision is the field of artificial intelligence that teaches computers to analyse images and video: to find objects, recognise faces, text and motion, and build a picture of the space around them.

Computer vision finds objects in an image

In plain words

On its own, a camera sees only millions of coloured dots - like a mosaic viewed up close. Computer vision is the ability to take a step back and say: "Aha, that is a cat on the sofa, and over there is the remote." A person does this instantly; a machine had to look through millions of labelled photos to learn it.

How it works

The camera passes on the image as a table of numbers - the brightness and colour of every pixel.
A neural network first looks for simple features: edges, corners, textures.
Then more complex ones: eyes, wheels, letters, silhouettes.
It works out what is in the image, exactly where and, if needed, how it is moving.
Other programs use the result: the camera, a robot, a car, search.

From a mosaic of pixels to a recognised object

Where you'll see it

Face unlock, sorting gallery photos by people and objects, recognising text in photos, Circle to Search, portrait mode, robot vacuums with cameras, driver assistance systems, security cameras and self-checkout in shops.

Does it matter when buying

You use it every day without noticing. When choosing a device, you see it in the quality of portrait mode, the speed of face unlock and how smart a robot vacuum is (does it steer around socks and cables?). And for privacy, remember that cameras with face recognition are becoming more common.

Myths

Myth: Face unlock is equally secure on every phone. Reality: No. A regular front camera is less reliable than 3D face scanning with an infrared sensor.
Myth: Computer vision sees just like a human. Reality: It can be "fooled" by strange lighting, stickers or an unusual angle.

FAQ

Does my gallery recognise faces offline?

On many phones, yes - the analysis happens on the device.

How does a robot vacuum see socks?

Through a camera and a neural network trained to recognise common obstacles.

Can I turn off face recognition in my gallery?

Usually yes, in the app settings.